tools
Saaras
Saaras is Sarvam AI's speech-to-text model for transcribing, translating, and processing speech in Indian languages.
In inglese
Saaras converts speech to text through REST, WebSocket, and batch APIs. It supports transcription, translation, verbatim output, transliteration, and code-mixed speech across Indian languages and English.
It is designed for developers building voice assistants, call-center tools, captioning systems, and multilingual applications. Usage is billed by audio duration; batch diarization costs more. It is not presented as a standalone consumer chat app.
Features
- Speech-to-text for 22 Indian languages and English
- Transcribe, translate, verbatim, translit, and codemix output modes
- REST, WebSocket, and batch processing endpoints
- Streaming speech recognition
- Batch transcription for longer audio files
- Batch diarization with speaker identification
- Supports code-mixed and regional speech
Use cases
- Build multilingual voice assistants
- Transcribe customer-support and call-center recordings
- Generate captions for Indian-language audio and video
- Translate spoken content between supported languages
- Analyze speakers in recorded conversations
Pros
Cons
Latest updates
- Sarvam Vision 2.1: Pushing the Pareto frontier of document intelligence (2.1)
Sarvam Vision 2.1 adds structured extraction and Indic handwritten recognition.
- Introducing Saaras V4 (V4)
Introduced Saaras V4 for multilingual ASR.
- Everything we announced at Sarvam Epoch
Announced new models, Sarvam Inference, Indus agents, Sarvam Code, and more.
- Introducing Sarvam Circle
Introduced Sarvam Circle to bring India's frontier AI to every institution.
- Open-Sourcing Sarvam 30B and 105B
Open-sourced Sarvam 30B and 105B.
Capabilities
- Speech to text — “From the same audio input, it can produce five transcript modes: verbatim, normalized, code-mixed, transliterated, and translated.” source
- Many languages — “Saaras V4 achieved SOTA performance on all 22 Indian languages and reaffirms our commitment to even the most low-resource Indian languages” source
- API — “Integrate Saaras v4 into your application with a simple API key from the Sarvam AI dashboard” source
Get it
Security
- SOC 2 Type II — “Complete data residency in India. ISO 27001 and SOC 2 Type II certified.” source
- SOC 2 — “Complete data residency in India. ISO 27001 and SOC 2 Type II certified.” source
- ISO 27001 — “We're ISO 27001:2022 certified and hold SOC 2 Type II.” source
- Not trained on your data — “Customer data is never used to train models for other customers” source
Pricing
- Prices checked
- 2026-09-25
Starter (Pay as you go)
- ₹3.00 per 1,000 characters (Text to speech, Real-time)
- ₹3.00 per 1,000 characters (Text to speech, Streaming)
- ₹30.00 per hour (Speech to text, Real-time)
- ₹30.00 per hour (Speech to text, Streaming)
- ₹30.00 per hour (Speech to text, Batch)
- ₹45.00 per hour (Speech to text, Batch with diarization)
- ₹128.10 per 1M tokens (GLM 5.2, input)
- ₹23.79 per 1M tokens (GLM 5.2, cached input)
- ₹402.60 per 1M tokens (GLM 5.2, output)
- ₹36.60 per 1M tokens (Gemma-4 31B, input)
- ₹13.73 per 1M tokens (Gemma-4 31B, cached input)
- ₹91.50 per 1M tokens (Gemma-4 31B, output)
- ₹29.28 per 1M tokens (Sarvam 105B, input)
- ₹10.98 per 1M tokens (Sarvam 105B, cached input)
- ₹73.20 per 1M tokens (Sarvam 105B, output)
- ₹29.28 per 1M tokens (Sarvam 105B Chat, input)
- ₹10.98 per 1M tokens (Sarvam 105B Chat, cached input)
- ₹73.20 per 1M tokens (Sarvam 105B Chat, output)
- ₹29.28 per 1M tokens (Sarvam 105B Conversations, input)
- ₹10.98 per 1M tokens (Sarvam 105B Conversations, cached input)
- ₹73.20 per 1M tokens (Sarvam 105B Conversations, output)
- ₹0.005 per character (Translate & transliterate, Pay as you go)
- ₹0.50 per page (Digitisation API)
- ₹1.00 per page (Extraction API)
- ₹1.67 per second (Dubbing API, Pay as you go)
- ₹0.005 (Translate & transliterate, Starter)
- ₹1.67 (Dubbing API, Starter)
- Start with free credits, then add exactly what you need
- Universal credits across supported APIs
- Credits never expire
- No monthly commitment
Enterprise
Price on request
- Tailored limits, throughput, pricing, and support
- Higher rate limits on every API
- Custom concurrency & throughput
- Volume discounts on credits
- Dedicated support & SLAs
Pro
- ₹0.005 per character (Translate & transliterate, Pro)
- ₹1.67 per second (Dubbing API, Pro)
Growth
- ₹0.0045 per character (Translate & transliterate, Growth)
- ₹1.25 per second (Dubbing API, Growth)
Business
- ₹0.004 per character (Translate & transliterate, Business)
- ₹1.20 per second (Dubbing API, Business)